There have been many attempts to solve the problem of continuous documentation. Tools like Swagger or Sphinx-autodoc can generate docs from OpenAPI specs or code annotations, but they only work in very narrow settings. They don't help much with higher-level docs like user guides, quickstarts, or example scripts. And they still rely on developers to manually maintain intermediate elements like docstrings.

With LLMs, updating high-level docs has gotten easier. You can prompt your AI IDE (e.g. Cursor) to update docs, and it usually works. But this approach has a few issues:

- You have to remember to prompt after every code change. - You don't know exactly which files are being used in context—either you specify them manually or trust the agent. - If teammates start updating docs separately in their own AI IDEs, things get messy fast. - [Minor] The updated docs often lose the structure or style of the original unless you're really careful with your prompts.

So I wanted something that would keep code and docs in sync automatically, with minimal human effort. So I built DeepDocs, an LLM-powered GitHub app for continuous documentation.

Once installed and configured, DeepDocs listens for code commits, detects relevant doc drift, and proactively opens a clean branch with suggested updates. Because it runs upstream, it only triggers after all necessary code changes are merged.

It keeps updates to the absolute necessary and respects the original formatting and structure. You don't need to prompt it or manually specify any file context, as DeepDocs always keeps the source code and docs in context.

The app is completely free to try on GitHub. I'd really appreciate any feedback or suggestions.